Machine learning-based prediction of low oil saturation sandstone reservoir parameters: a case study of Lower Karamay Formation in Xia 77 well block of Xiazijie Oilfield, Junggar Basin

The Lower Karamay Formation in the Xia 77 well block of the Xiazijie Oilfield in the Junggar Basin features a complex oil and water relationship in its ultra-low porosity and ultra-low permeability reservoirs. These reservoirs are characterized by low production, high water content, low oil saturation, poor correlation between porosity and permeability, unclear relationship between reservoir parameters and logging responses and difficult identification of oil and water layers. Conventional methods for evaluating and predicting reservoir parameters are poorly suited for this block. Through the analysis of lithology, physical properties and oil-bearing characteristics, it was determined that the reservoir lithology of the Lower Karamay Formation is dominated by glutenite and gravelly sandstones, with mixed-layers of illite and smectite as the dominant clay mineral. The reservoirs are characterized by low porosity and ultra-low permeability with primary intergranular and residual intergranular pores as the main storage space. By establishing an oil saturation interpretation model, it was confirmed that the reservoirs in this area are low oil saturation reservoirs, with oil saturation generally ranging between 36%-55%. The physical properties and oil content of glutenite reservoirs are superior to those of medium to fine sandstones, with reservoir physical properties controlling oil content and exhibiting low saturation characteristics. Electrical properties are influenced by both oil content and lithology. Through studying the formation mechanism of low oil saturation oil reservoirs, it was found that the microscopic pore structure of the reservoirs is the main cause of low oil saturation. By selecting sensitive parameters and utilizing data from natural gamma, resistivity, and acoustic time difference logging, BP neural network technology based on machine learning was introduced to calculate and predict porosity, permeability, and water saturation for the Lower Karamay Formation in Xia 77 well block. The prediction accuracy of reservoir parameters exceeded 80%. The conclusions and methods derived from this study can provide a basis and reference for the prediction of physical parameters in low oil saturation tight sandstone reservoirs.